基于S-HOG+C算子的变电作业人员着装分析方法研究

Research on Dressing Analysis Method of Substation Workers Based on S-HOG+C Operator

为了在变电作业复杂环境中采用HOG和HOC特征进行着装分析时能满足识别精确度的要求,研究了对HOG+C算子的改进方法。首先,考虑到变电作业人员的头盔、上衣和下衣3个部分的形状特征明显不同,且各部分之间差异较大,提出对作业人员图像进行结构划分,然后再对划分所得的3个单元格分别同时提取HOG和HOC特征来训练线性SVM分类器,通过设计S-HOG+C算子来提高变电作业人员着装分析系统的识别精准度,最后采用3个评估指标分析了S-HOG+C算子在变电作业人员着装分析方面的性能。结果表明,在识别准确度方面S-HOG+C算子优于HOG+C算子。

In order to satisfy the requirement of recognition accuracy when using HOG and HOC features for dress analysis in complex environment of substation operation,an improved method of HOG+C operator was studied.Firstly,considering that the shape characteristics of three parts of helmet,jacket and jacket of substation operators were obviously different,and there were great differences among them,a structural division of operator image was proposed.Then,HOG and HOC features were extracted from the three cells to train the linear SVM classifier at the same time.Through the design,the linear SVM classifier was trained.S-HOG+C operator was used to improve the identification accuracy of substation operator's dress analysis system.Finally,the performance of S-HOG+C operator in substation operator's dress analysis was analyzed by three evaluation indexes.The results show that S-HOG+C operator is superior to HOG+C operator in recognition accuracy.